fix: Improve agent subprocess error handling (#489)

...

Author: charlesvien

apps/array/src/renderer/features/sessions/components/SessionView.tsx

@@ -7,7 +7,8 @@ import {
   useSessionActions,
 } from "@features/sessions/stores/sessionStore";
 import type { Plan } from "@features/sessions/types";
-import { Box, ContextMenu, Flex } from "@radix-ui/themes";
+import { Warning } from "@phosphor-icons/react";
+import { Box, Button, ContextMenu, Flex, Text } from "@radix-ui/themes";
 import {
   type AcpMessage,
   isJsonRpcNotification,
@@ -42,8 +43,15 @@ interface SessionViewProps {
   onCancelPrompt: () => void;
   repoPath?: string | null;
   isCloud?: boolean;
+  hasError?: boolean;
+  errorMessage?: string;
+  onRetry?: () => void;
+  onDelete?: () => void;
 }
 
+const DEFAULT_ERROR_MESSAGE =
+  "Failed to resume this session. The working directory may have been deleted. Please start a new task.";
+
 export function SessionView({
   events,
   taskId,
@@ -54,6 +62,10 @@ export function SessionView({
   onCancelPrompt,
   repoPath,
   isCloud = false,
+  hasError = false,
+  errorMessage = DEFAULT_ERROR_MESSAGE,
+  onRetry,
+  onDelete,
 }: SessionViewProps) {
   const showRawLogs = useShowRawLogs();
   const { setShowRawLogs } = useSessionViewActions();
@@ -233,7 +245,44 @@ export function SessionView({
 
           <PlanStatusBar plan={latestPlan} />
 
-          {firstPendingPermission ? (
+          {hasError ? (
+            <Flex
+              align="center"
+              justify="center"
+              direction="column"
+              gap="2"
+              className="absolute inset-0"
+            >
+              <Warning size={32} weight="duotone" color="var(--red-9)" />
+              <Text size="3" weight="medium" color="red">
+                Session Error
+              </Text>
+              <Text
+                size="2"
+                align="center"
+                className="max-w-md px-4 text-gray-11"
+              >
+                {errorMessage}
+              </Text>
+              <Flex gap="2" mt="2">
+                {onRetry && (
+                  <Button variant="soft" size="2" onClick={onRetry}>
+                    Retry
+                  </Button>
+                )}
+                {onDelete && (
+                  <Button
+                    variant="soft"
+                    size="2"
+                    color="red"
+                    onClick={onDelete}
+                  >
+                    Delete Task
+                  </Button>
+                )}
+              </Flex>
+            </Flex>
+          ) : firstPendingPermission ? (
             <InlinePermissionSelector
               title={firstPendingPermission.title}
               options={firstPendingPermission.options}

apps/array/src/renderer/features/sessions/stores/sessionStore.ts

@@ -46,6 +46,7 @@ export interface AgentSession {
   events: AcpMessage[];
   startedAt: number;
   status: "connecting" | "connected" | "disconnected" | "error";
+  errorMessage?: string;
   isPromptPending: boolean;
   isCloud: boolean;
   logUrl?: string;
@@ -91,6 +92,7 @@ interface SessionActions {
     customInput?: string,
   ) => Promise<void>;
   cancelPermission: (taskId: string, toolCallId: string) => Promise<void>;
+  clearSessionError: (taskId: string) => void;
 }
 
 interface AuthCredentials {
@@ -160,6 +162,14 @@ function subscribeToChannel(taskRunId: string) {
       },
       onError: (err) => {
         log.error("Session subscription error", { taskRunId, error: err });
+        useStore.setState((state) => {
+          const session = state.sessions[taskRunId];
+          if (session) {
+            session.status = "error";
+            session.errorMessage =
+              "Lost connection to the agent. Please restart the task.";
+          }
+        });
       },
     },
   );
@@ -640,7 +650,11 @@ const useStore = create<SessionStore>()(
         }
       } else {
         unsubscribeFromChannel(taskRunId);
-        removeSession(taskRunId);
+        updateSession(taskRunId, {
+          status: "error",
+          errorMessage:
+            "Failed to reconnect to the agent. Please restart the task.",
+        });
       }
     };
 
@@ -652,14 +666,14 @@ const useStore = create<SessionStore>()(
       executionMode?: "plan" | "acceptEdits",
     ) => {
       if (!auth.client) {
-        log.error("API client not available");
-        return;
+        throw new Error(
+          "Unable to reach server. Please check your connection.",
+        );
       }
 
       const taskRun = await auth.client.createTaskRun(taskId);
       if (!taskRun?.id) {
-        log.error("Task run created without ID");
-        return;
+        throw new Error("Failed to create task run. Please try again.");
       }
 
       const persistedMode = getPersistedTaskMode(taskId);
@@ -776,6 +790,12 @@ const useStore = create<SessionStore>()(
             const auth = getAuthCredentials();
             if (!auth) {
               log.error("Missing auth credentials");
+              const taskRunId = latestRun?.id ?? `error-${taskId}`;
+              const session = createBaseSession(taskRunId, taskId, isCloud);
+              session.status = "error";
+              session.errorMessage =
+                "Authentication required. Please sign in to continue.";
+              addSession(session);
               return;
             }
 
@@ -787,6 +807,32 @@ const useStore = create<SessionStore>()(
                 taskDescription,
               );
             } else if (latestRun?.id && latestRun?.log_url) {
+              // Check if workspace still exists before attempting reconnect
+              const workspaceExists = await trpcVanilla.workspace.verify.query({
+                taskId,
+              });
+
+              if (!workspaceExists) {
+                // Workspace was deleted - show historical logs in error state
+                log.warn("Workspace no longer exists, showing error state", {
+                  taskId,
+                });
+                const { rawEntries } = await fetchSessionLogs(
+                  latestRun.log_url,
+                );
+                const events = convertStoredEntriesToEvents(rawEntries);
+
+                const session = createBaseSession(latestRun.id, taskId, false);
+                session.events = events;
+                session.logUrl = latestRun.log_url;
+                session.status = "error";
+                session.errorMessage =
+                  "The working directory for this task no longer exists. Please start a new task.";
+
+                addSession(session);
+                return;
+              }
+
               await reconnectToLocalSession(
                 taskId,
                 latestRun.id,
@@ -807,6 +853,27 @@ const useStore = create<SessionStore>()(
             const message =
               error instanceof Error ? error.message : String(error);
             log.error("Failed to connect to task", { message });
+
+            // Create session in error state so user sees what happened
+            const taskRunId = latestRun?.id ?? `error-${taskId}`;
+            const session = createBaseSession(taskRunId, taskId, isCloud);
+            session.status = "error";
+            session.errorMessage = `Failed to connect to the agent: ${message}`;
+
+            // Try to load historical logs if available
+            if (latestRun?.log_url) {
+              try {
+                const { rawEntries } = await fetchSessionLogs(
+                  latestRun.log_url,
+                );
+                session.events = convertStoredEntriesToEvents(rawEntries);
+                session.logUrl = latestRun.log_url;
+              } catch {
+                // Ignore log fetch errors - just show error state without logs
+              }
+            }
+
+            addSession(session);
           } finally {
             connectAttempts.delete(taskId);
           }
@@ -1106,6 +1173,14 @@ const useStore = create<SessionStore>()(
             });
           }
         },
+
+        clearSessionError: (taskId: string) => {
+          const session = getSessionByTaskId(taskId);
+          if (session) {
+            removeSession(session.taskRunId);
+          }
+          connectAttempts.delete(taskId);
+        },
       },
     };
   }),

apps/array/src/renderer/features/task-detail/components/TaskLogsPanel.tsx

@@ -8,6 +8,7 @@ import {
 import { useSettingsStore } from "@features/settings/stores/settingsStore";
 import { useTaskViewedStore } from "@features/sidebar/stores/taskViewedStore";
 import { useTaskData } from "@features/task-detail/hooks/useTaskData";
+import { useDeleteTask } from "@features/tasks/hooks/useTasks";
 import {
   selectWorktreePath,
   useWorkspaceStore,
@@ -32,13 +33,17 @@ export function TaskLogsPanel({ taskId, task }: TaskLogsPanelProps) {
   const repoPath = worktreePath ?? taskData.repoPath;
 
   const session = useSessionForTask(taskId);
-  const { connectToTask, sendPrompt, cancelPrompt } = useSessionActions();
+  const { connectToTask, sendPrompt, cancelPrompt, clearSessionError } =
+    useSessionActions();
+  const { deleteWithConfirm } = useDeleteTask();
   const markActivity = useTaskViewedStore((state) => state.markActivity);
   const markAsViewed = useTaskViewedStore((state) => state.markAsViewed);
   const requestFocus = useDraftStore((s) => s.actions.requestFocus);
 
   const isRunning =
     session?.status === "connected" || session?.status === "connecting";
+  const hasError = session?.status === "error";
+  const errorMessage = session?.errorMessage;
 
   const events = session?.events ?? [];
   const isPromptPending = session?.isPromptPending ?? false;
@@ -54,7 +59,12 @@ export function TaskLogsPanel({ taskId, task }: TaskLogsPanelProps) {
     if (!repoPath) return;
     if (isConnecting.current) return;
 
-    if (session?.status === "connected" || session?.status === "connecting") {
+    // Don't reconnect if already connected, connecting, or in error state
+    if (
+      session?.status === "connected" ||
+      session?.status === "connecting" ||
+      session?.status === "error"
+    ) {
       return;
     }
 
@@ -122,6 +132,21 @@ export function TaskLogsPanel({ taskId, task }: TaskLogsPanelProps) {
 
   const { appendUserShellExecute } = useSessionActions();
 
+  const handleRetry = useCallback(() => {
+    if (!repoPath) return;
+    clearSessionError(taskId);
+    connectToTask({ task, repoPath });
+  }, [taskId, repoPath, task, clearSessionError, connectToTask]);
+
+  const handleDelete = useCallback(() => {
+    const hasWorktree = !!worktreePath;
+    deleteWithConfirm({
+      taskId,
+      taskTitle: task.title ?? task.description ?? "Untitled",
+      hasWorktree,
+    });
+  }, [taskId, task, worktreePath, deleteWithConfirm]);
+
   const handleBashCommand = useCallback(
     async (command: string) => {
       if (!repoPath) return;
@@ -152,6 +177,10 @@ export function TaskLogsPanel({ taskId, task }: TaskLogsPanelProps) {
           onCancelPrompt={handleCancelPrompt}
           repoPath={repoPath}
           isCloud={session?.isCloud ?? false}
+          hasError={hasError}
+          errorMessage={errorMessage}
+          onRetry={handleRetry}
+          onDelete={handleDelete}
         />
       </Box>
     </BackgroundWrapper>